Human behaviour in fire and emergency evacuation design

BRE has a proven and internationally recognised expertise in this field.

It has been at the forefront of the application of research to the development of regulatory tools and systems. 

Through our extensive research programmes into human behaviour in fire incidents and experimental evacuation studies, we have developed a unique knowledge and understanding of the factors controlling occupant escape behaviour (such as alarm systems, fire-safety management, occupancy type and building complexity) important for the development of effective evacuation strategies.

We use this knowledge base to develop methods for the quantification and modelling of human behaviour during emergencies, which have been incorporated into engineering calculation and computational tools. We understand the impact on physical means of escape provisions, which constitute one of the greatest restraints and cost elements in building design and construction.

Safe means of escape in fire and other emergencies is a principal requirement for design and operation of buildings and transport systems.

The time required for escape (known as RSET) is an essential part of performance-based fire safety design. Once a general warning has been given to the building occupants, evacuation commences with a series of initial behaviours (the pre-movement time) followed by movement into and through escape routes (the travel time). Also important are the interactions between pre-movement and travel behaviours for all individual occupants.

Numerical models of the evacuation process vary widely in their degrees of sophistication. The simplest treat the population as a homogeneous fluid, or mindless particles, and concentrate on the flow capacity of the building. At the other extreme there are detailed simulations where each person is treated individually, with explicit behavioural rules.

Building geometry can either be described by a coarse network (each node ~ 1 room) or a fine network (each node ~ 0.5m "pixel"). Fine networks are necessary to calculate complex flow rates from first principles, coarse networks would require empirical equations. The more "homogeneous" the population behaviour, the less distinct the difference between coarse and fine networks (ie all flows heading to exits, no contraflows & limited merging). Human behaviour is the most complex and difficult aspect to simulate, yet is crucial to accurate results.

BRE has developed two egress simulations, GridFlow and CRISP. Both models treat each person individually and use a fine network to regulate the movement process.

GridFlow has a simplified model of human behaviour, where all the "pre-movement" activities a represented by a single time delay (which may differ for each person). When these activities are complete, people move to the outside, via the nearest, preferred, or a randomly-chosen exit. Despite this simplified behaviour, there are many situations where the model can predict RSET with reasonable accuracy.

CRISP is a simulation of entire fire scenarios, developed as a risk assessment tool using Monte-Carlo methods. However, it can also be run in a stand-alone evacuation mode, without simulating the fire or calculating toxic exposures. The behaviour of the occupants is much more detailed than in GridFlow, making the CRISP model applicable to a wider range of scenarios.

Their behaviour can be described in terms of actions, which may be abandoned, and substituted by new ones, depending on the state of the environment. Rational decisions are made based on current knowledge (which may be limited and/or incorrect). People may move from one room to another in the building, not just to the outside.

The model attempts to calculate "pre-movement time" (rather than use an empirical distribution) in terms of the time delays associated with various actions performed by the occupants in response to the early fire cues.

The occupants may perform a number of actions (e.g. investigate, warn others) before actually starting to escape (thus the term "pre-movement time" is not strictly accurate). If the occupants 'pre-movement' actions do not actually require him to move, then all these actions can be merged into a single delay in reacting to the alarm (as in GridFlow).

In "evacuation mode", all the building occupants are assumed to be "alerted" at the start of the simulation, unless a phased evacuation is being simulated. Once they have finished reacting, their behavioural rules govern what they do next. Some of these may be quite complex e.g. members of staff may be required to search a building and order other occupants to leave.

In "risk assessment" mode, as the people move around, they are exposed to smoke and acquire a Fractional Effective Dose (FED). When the FED reaches 100%, the person is assumed to be "dead". The risk is expressed simply in terms of the fraction of people originally present who end up "dead", averaged over a sufficiently large Monte-Carlo sample.

For "evacuation mode", the model also calculates detailed statistics of the evacuation process. At the most basic level, there is the time required to clear a given proportion of the initial population from the building.

For both CRISP and GridFlow, it is possible to create pictures of the evacuation process, either by taking a number of 'snapshots' at regular intervals, or 3D animations can also be created.